model-usage

What This Skill Does

The model-usage skill provides a sophisticated interface for interacting with the CodexBar CLI to retrieve, parse, and analyze cost data associated with your LLM usage. By leveraging local cost logs, it allows users to gain granular insights into their spending habits across different providers like Codex and Claude. The skill automates the extraction of complex JSON logs and translates them into actionable, human-readable summaries. It supports both high-level overviews of total consumption and deep dives into specific model performance, helping you maintain budget discipline without manually parsing log files. Whether you are tracking daily spikes or performing a monthly audit, this skill acts as the bridge between raw data exports and strategic financial decision-making.

Installation

To install this skill, use the ClawHub CLI tool integrated into OpenClaw. Run the following command in your terminal:

clawhub install openclaw/skills/skills/steipete/model-usage

Once installed, ensure you have the CodexBar CLI configured in your system path. If you are operating on a Linux environment, please verify the documentation in references/codexbar-cli.md for specific pathing requirements. The Python environment must be active to allow the bundled script to process the JSON inputs correctly.

Use Cases

This skill is designed for power users, developers, and project managers who rely heavily on AI assistance. Key use cases include:

• Cost Auditing: Quickly identifying which models are contributing most to your monthly bill.
• Trend Analysis: Comparing current model performance versus previous iterations by analyzing historical daily logs.
• Automation Reporting: Integrating cost summaries into automated daily check-ins or project status dashboards.
• Resource Allocation: Determining whether to scale back usage on high-cost models or optimize prompts to reduce token counts.

Example Prompts

1. "Check my CodexBar usage for the current model and tell me what the cost breakdown looks like for today."
2. "Show me a summary of all Claude model costs from my latest CLI export in JSON format."
3. "Summarize the total cost for all models over the entire history of my CodexBar logs."

Tips & Limitations

• Data Limitations: The skill relies on CodexBar logs. If the logs are truncated or corrupted, the summary will reflect the incomplete data.
• Performance: While the script is optimized for quick local execution, processing massive JSON files containing years of history may result in slight latency.
• Tokens vs. Cost: Note that the tool currently reports on monetary costs only. Token counts are not explicitly separated by model in the standard CodexBar output, so you should rely on the cost-per-model metrics as your primary indicator of usage impact.
• Pro-Tip: Use the --format json --pretty flags when piping output to other tools for seamless integration with downstream data analysis applications.